2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.74 2008/11/13 02:18:43 dillon Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/vnode.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/nlookup.h>
44 #include <sys/fcntl.h>
45 #include <sys/sysctl.h>
51 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them
54 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
56 int hammer_debug_general;
57 int hammer_debug_debug = 1; /* medium-error panics */
58 int hammer_debug_inode;
59 int hammer_debug_locks;
60 int hammer_debug_btree;
62 int hammer_debug_recover; /* -1 will disable, +1 will force */
63 int hammer_debug_recover_faults;
64 int hammer_debug_critical; /* non-zero enter debugger on error */
65 int hammer_cluster_enable = 1; /* enable read clustering by default */
66 int hammer_live_dedup = 0;
67 int hammer_count_fsyncs;
68 int hammer_count_inodes;
69 int hammer_count_iqueued;
70 int hammer_count_reclaiming;
71 int hammer_count_records;
72 int hammer_count_record_datas;
73 int hammer_count_volumes;
74 int hammer_count_buffers;
75 int hammer_count_nodes;
76 int64_t hammer_count_extra_space_used;
77 int64_t hammer_stats_btree_lookups;
78 int64_t hammer_stats_btree_searches;
79 int64_t hammer_stats_btree_inserts;
80 int64_t hammer_stats_btree_deletes;
81 int64_t hammer_stats_btree_elements;
82 int64_t hammer_stats_btree_splits;
83 int64_t hammer_stats_btree_iterations;
84 int64_t hammer_stats_btree_root_iterations;
85 int64_t hammer_stats_record_iterations;
87 int64_t hammer_stats_file_read;
88 int64_t hammer_stats_file_write;
89 int64_t hammer_stats_file_iopsr;
90 int64_t hammer_stats_file_iopsw;
91 int64_t hammer_stats_disk_read;
92 int64_t hammer_stats_disk_write;
93 int64_t hammer_stats_inode_flushes;
94 int64_t hammer_stats_commits;
95 int64_t hammer_stats_undo;
96 int64_t hammer_stats_redo;
98 int hammer_count_dirtybufspace; /* global */
99 int hammer_count_refedbufs; /* global */
100 int hammer_count_reservations;
101 int hammer_count_io_running_read;
102 int hammer_count_io_running_write;
103 int hammer_count_io_locked;
104 int hammer_limit_dirtybufspace; /* per-mount */
105 int hammer_limit_running_io; /* per-mount */
106 int hammer_limit_recs; /* as a whole XXX */
107 int hammer_limit_inode_recs = 2048; /* per inode */
108 int hammer_limit_reclaim;
109 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
110 int hammer_limit_redo = 4096 * 1024; /* per inode */
111 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */
112 int hammer_bio_count;
113 int hammer_verify_zone;
114 int hammer_verify_data = 1;
115 int hammer_write_mode;
116 int hammer_double_buffer;
117 int hammer_yield_check = 16;
118 int hammer_fsync_mode = 3;
119 int64_t hammer_contention_count;
120 int64_t hammer_zone_limit;
123 * Live dedup debug counters (sysctls are writable so that counters
124 * can be reset from userspace).
126 int64_t hammer_live_dedup_vnode_bcmps = 0;
127 int64_t hammer_live_dedup_device_bcmps = 0;
128 int64_t hammer_live_dedup_findblk_failures = 0;
129 int64_t hammer_live_dedup_bmap_saves = 0;
132 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
135 &hammer_supported_version, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
137 &hammer_debug_general, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
139 &hammer_debug_io, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
141 &hammer_debug_debug, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
143 &hammer_debug_inode, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
145 &hammer_debug_locks, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
147 &hammer_debug_btree, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
149 &hammer_debug_tid, 0, "");
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
151 &hammer_debug_recover, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
153 &hammer_debug_recover_faults, 0, "");
154 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
155 &hammer_debug_critical, 0, "");
156 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
157 &hammer_cluster_enable, 0, "");
159 * 0 - live dedup is disabled
160 * 1 - dedup cache is populated on reads only
161 * 2 - dedup cache is populated on both reads and writes
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RW,
164 &hammer_live_dedup, 0, "Enable live dedup");
166 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
167 &hammer_limit_dirtybufspace, 0, "");
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_running_io, CTLFLAG_RW,
169 &hammer_limit_running_io, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
171 &hammer_limit_recs, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
173 &hammer_limit_inode_recs, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaim, CTLFLAG_RW,
175 &hammer_limit_reclaim, 0, "");
176 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
177 &hammer_live_dedup_cache_size, 0,
178 "Number of cache entries");
179 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
180 &hammer_limit_redo, 0, "");
182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
183 &hammer_count_fsyncs, 0, "");
184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
185 &hammer_count_inodes, 0, "");
186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
187 &hammer_count_iqueued, 0, "");
188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
189 &hammer_count_reclaiming, 0, "");
190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
191 &hammer_count_records, 0, "");
192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
193 &hammer_count_record_datas, 0, "");
194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
195 &hammer_count_volumes, 0, "");
196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
197 &hammer_count_buffers, 0, "");
198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
199 &hammer_count_nodes, 0, "");
200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
201 &hammer_count_extra_space_used, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
204 &hammer_stats_btree_searches, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
206 &hammer_stats_btree_lookups, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
208 &hammer_stats_btree_inserts, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
210 &hammer_stats_btree_deletes, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
212 &hammer_stats_btree_elements, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
214 &hammer_stats_btree_splits, 0, "");
215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
216 &hammer_stats_btree_iterations, 0, "");
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
218 &hammer_stats_btree_root_iterations, 0, "");
219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
220 &hammer_stats_record_iterations, 0, "");
222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
223 &hammer_stats_file_read, 0, "");
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
225 &hammer_stats_file_write, 0, "");
226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
227 &hammer_stats_file_iopsr, 0, "");
228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
229 &hammer_stats_file_iopsw, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
231 &hammer_stats_disk_read, 0, "");
232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
233 &hammer_stats_disk_write, 0, "");
234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
235 &hammer_stats_inode_flushes, 0, "");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
237 &hammer_stats_commits, 0, "");
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
239 &hammer_stats_undo, 0, "");
240 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
241 &hammer_stats_redo, 0, "");
243 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
244 &hammer_live_dedup_vnode_bcmps, 0,
245 "successful vnode buffer comparisons");
246 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
247 &hammer_live_dedup_device_bcmps, 0,
248 "successful device buffer comparisons");
249 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
250 &hammer_live_dedup_findblk_failures, 0,
251 "block lookup failures for comparison");
252 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
253 &hammer_live_dedup_bmap_saves, 0,
254 "useful physical block lookups");
256 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
257 &hammer_count_dirtybufspace, 0, "");
258 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
259 &hammer_count_refedbufs, 0, "");
260 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
261 &hammer_count_reservations, 0, "");
262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
263 &hammer_count_io_running_read, 0, "");
264 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
265 &hammer_count_io_locked, 0, "");
266 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
267 &hammer_count_io_running_write, 0, "");
268 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
269 &hammer_zone_limit, 0, "");
270 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
271 &hammer_contention_count, 0, "");
272 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
273 &hammer_autoflush, 0, "");
274 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
275 &hammer_verify_zone, 0, "");
276 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
277 &hammer_verify_data, 0, "");
278 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
279 &hammer_write_mode, 0, "");
280 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
281 &hammer_double_buffer, 0, "");
282 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
283 &hammer_yield_check, 0, "");
284 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
285 &hammer_fsync_mode, 0, "");
287 KTR_INFO_MASTER(hammer);
292 static void hammer_free_hmp(struct mount *mp);
294 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
296 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
297 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
298 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
300 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
302 static int hammer_vfs_sync(struct mount *mp, int waitfor);
303 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
304 ino_t ino, struct vnode **vpp);
305 static int hammer_vfs_init(struct vfsconf *conf);
306 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
307 struct fid *fhp, struct vnode **vpp);
308 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
309 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
310 int *exflagsp, struct ucred **credanonp);
313 static struct vfsops hammer_vfsops = {
314 .vfs_mount = hammer_vfs_mount,
315 .vfs_unmount = hammer_vfs_unmount,
316 .vfs_root = hammer_vfs_root,
317 .vfs_statfs = hammer_vfs_statfs,
318 .vfs_statvfs = hammer_vfs_statvfs,
319 .vfs_sync = hammer_vfs_sync,
320 .vfs_vget = hammer_vfs_vget,
321 .vfs_init = hammer_vfs_init,
322 .vfs_vptofh = hammer_vfs_vptofh,
323 .vfs_fhtovp = hammer_vfs_fhtovp,
324 .vfs_checkexp = hammer_vfs_checkexp
327 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
329 VFS_SET(hammer_vfsops, hammer, 0);
330 MODULE_VERSION(hammer, 1);
333 hammer_vfs_init(struct vfsconf *conf)
337 if (hammer_limit_recs == 0) {
338 hammer_limit_recs = nbuf * 25;
339 n = kmalloc_limit(M_HAMMER) / 512;
340 if (hammer_limit_recs > n)
341 hammer_limit_recs = n;
343 if (hammer_limit_dirtybufspace == 0) {
344 hammer_limit_dirtybufspace = hidirtybufspace / 2;
345 if (hammer_limit_dirtybufspace < 100)
346 hammer_limit_dirtybufspace = 100;
350 * Set reasonable limits to maintain an I/O pipeline. This is
351 * used by the flush code which explicitly initiates I/O, and
354 * The system-driven buffer cache uses vfs.lorunningspace and
355 * vfs.hirunningspace globally.
357 if (hammer_limit_running_io == 0)
358 hammer_limit_running_io = hammer_limit_dirtybufspace;
359 if (hammer_limit_running_io > 10 * 1024 * 1024)
360 hammer_limit_running_io = 10 * 1024 * 1024;
363 * The hammer_inode structure detaches from the vnode on reclaim.
364 * This limits the number of inodes in this state to prevent a
365 * memory pool blowout.
367 if (hammer_limit_reclaim == 0)
368 hammer_limit_reclaim = desiredvnodes / 10;
374 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
377 struct hammer_mount_info info;
379 hammer_volume_t rootvol;
380 struct vnode *rootvp;
381 struct vnode *devvp = NULL;
382 const char *upath; /* volume name in userspace */
383 char *path; /* volume name in system space */
387 char *next_volume_ptr = NULL;
390 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
393 bzero(&info, sizeof(info));
398 next_volume_ptr = mp->mnt_stat.f_mntfromname;
400 /* Count number of volumes separated by ':' */
401 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
407 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
409 if ((error = copyin(data, &info, sizeof(info))) != 0)
414 * updating or new mount
416 if (mp->mnt_flag & MNT_UPDATE) {
417 hmp = (void *)mp->mnt_data;
418 KKASSERT(hmp != NULL);
420 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
426 * master-id validation. The master id may not be changed by a
429 if (info.hflags & HMNT_MASTERID) {
430 if (hmp && hmp->master_id != info.master_id) {
431 kprintf("hammer: cannot change master id "
432 "with mount update\n");
435 master_id = info.master_id;
436 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
440 master_id = hmp->master_id;
446 * Internal mount data structure
449 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
450 mp->mnt_data = (qaddr_t)hmp;
452 /*TAILQ_INIT(&hmp->recycle_list);*/
455 * Make sure kmalloc type limits are set appropriately.
457 * Our inode kmalloc group is sized based on maxvnodes
458 * (controlled by the system, not us).
460 kmalloc_create(&hmp->m_misc, "HAMMER-others");
461 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
463 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
465 hmp->root_btree_beg.localization = 0x00000000U;
466 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
467 hmp->root_btree_beg.key = -0x8000000000000000LL;
468 hmp->root_btree_beg.create_tid = 1;
469 hmp->root_btree_beg.delete_tid = 1;
470 hmp->root_btree_beg.rec_type = 0;
471 hmp->root_btree_beg.obj_type = 0;
473 hmp->root_btree_end.localization = 0xFFFFFFFFU;
474 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
475 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
476 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
477 hmp->root_btree_end.delete_tid = 0; /* special case */
478 hmp->root_btree_end.rec_type = 0xFFFFU;
479 hmp->root_btree_end.obj_type = 0;
481 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
482 hmp->krate.count = -16; /* initial burst */
484 hmp->sync_lock.refs = 1;
485 hmp->free_lock.refs = 1;
486 hmp->undo_lock.refs = 1;
487 hmp->blkmap_lock.refs = 1;
488 hmp->snapshot_lock.refs = 1;
489 hmp->volume_lock.refs = 1;
491 TAILQ_INIT(&hmp->delay_list);
492 TAILQ_INIT(&hmp->flush_group_list);
493 TAILQ_INIT(&hmp->objid_cache_list);
494 TAILQ_INIT(&hmp->undo_lru_list);
495 TAILQ_INIT(&hmp->reclaim_list);
497 RB_INIT(&hmp->rb_dedup_crc_root);
498 RB_INIT(&hmp->rb_dedup_off_root);
499 TAILQ_INIT(&hmp->dedup_lru_list);
501 hmp->hflags &= ~HMNT_USERFLAGS;
502 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
504 hmp->master_id = master_id;
507 mp->mnt_flag |= MNT_RDONLY;
508 hmp->asof = info.asof;
510 hmp->asof = HAMMER_MAX_TID;
513 hmp->volume_to_remove = -1;
516 * Re-open read-write if originally read-only, or vise-versa.
518 * When going from read-only to read-write execute the stage2
519 * recovery if it has not already been run.
521 if (mp->mnt_flag & MNT_UPDATE) {
522 lwkt_gettoken(&hmp->fs_token);
524 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
525 kprintf("HAMMER read-only -> read-write\n");
527 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
528 hammer_adjust_volume_mode, NULL);
529 rootvol = hammer_get_root_volume(hmp, &error);
531 hammer_recover_flush_buffers(hmp, rootvol, 1);
532 error = hammer_recover_stage2(hmp, rootvol);
533 bcopy(rootvol->ondisk->vol0_blockmap,
535 sizeof(hmp->blockmap));
536 hammer_rel_volume(rootvol, 0);
538 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
539 hammer_reload_inode, NULL);
540 /* kernel clears MNT_RDONLY */
541 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
542 kprintf("HAMMER read-write -> read-only\n");
543 hmp->ronly = 1; /* messy */
544 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
545 hammer_reload_inode, NULL);
547 hammer_flusher_sync(hmp);
548 hammer_flusher_sync(hmp);
549 hammer_flusher_sync(hmp);
551 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
552 hammer_adjust_volume_mode, NULL);
554 lwkt_reltoken(&hmp->fs_token);
558 RB_INIT(&hmp->rb_vols_root);
559 RB_INIT(&hmp->rb_inos_root);
560 RB_INIT(&hmp->rb_redo_root);
561 RB_INIT(&hmp->rb_nods_root);
562 RB_INIT(&hmp->rb_undo_root);
563 RB_INIT(&hmp->rb_resv_root);
564 RB_INIT(&hmp->rb_bufs_root);
565 RB_INIT(&hmp->rb_pfsm_root);
567 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
569 RB_INIT(&hmp->volu_root);
570 RB_INIT(&hmp->undo_root);
571 RB_INIT(&hmp->data_root);
572 RB_INIT(&hmp->meta_root);
573 RB_INIT(&hmp->lose_root);
574 TAILQ_INIT(&hmp->iorun_list);
576 lwkt_token_init(&hmp->fs_token, "hammerfs");
577 lwkt_token_init(&hmp->io_token, "hammerio");
579 lwkt_gettoken(&hmp->fs_token);
584 path = objcache_get(namei_oc, M_WAITOK);
586 for (i = 0; i < info.nvolumes; ++i) {
591 KKASSERT(next_volume_ptr != NULL);
593 if (*next_volume_ptr != '/') {
595 strcpy(path, "/dev/");
598 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
599 if (*next_volume_ptr == '\0') {
601 } else if (*next_volume_ptr == ':') {
605 path[k] = *next_volume_ptr;
612 cdev_t dev = kgetdiskbyname(path);
613 error = bdevvp(dev, &devvp);
615 kprintf("hammer_mountroot: can't find devvp\n");
618 error = copyin(&info.volumes[i], &upath,
621 error = copyinstr(upath, path,
625 error = hammer_install_volume(hmp, path, devvp);
629 objcache_put(namei_oc, path);
632 * Make sure we found a root volume
634 if (error == 0 && hmp->rootvol == NULL) {
635 kprintf("hammer_mount: No root volume found!\n");
640 * Check that all required volumes are available
642 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
643 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
648 /* called with fs_token held */
654 * No errors, setup enough of the mount point so we can lookup the
657 mp->mnt_iosize_max = MAXPHYS;
658 mp->mnt_kern_flag |= MNTK_FSMID;
661 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
662 * will acquire a per-mount token prior to entry and release it
663 * on return, so even if we do not specify it we no longer get
664 * the BGL regardlless of how we are flagged.
666 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
667 /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
670 * note: f_iosize is used by vnode_pager_haspage() when constructing
673 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
674 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
676 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
677 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
679 mp->mnt_maxsymlinklen = 255;
680 mp->mnt_flag |= MNT_LOCAL;
682 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
683 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
684 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
687 * The root volume's ondisk pointer is only valid if we hold a
690 rootvol = hammer_get_root_volume(hmp, &error);
695 * Perform any necessary UNDO operations. The recovery code does
696 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
697 * and then re-copy it again after recovery is complete.
699 * If this is a read-only mount the UNDO information is retained
700 * in memory in the form of dirty buffer cache buffers, and not
701 * written back to the media.
703 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
704 sizeof(hmp->blockmap));
707 * Check filesystem version
709 hmp->version = rootvol->ondisk->vol_version;
710 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
711 hmp->version > HAMMER_VOL_VERSION_MAX) {
712 kprintf("HAMMER: mount unsupported fs version %d\n",
719 * The undo_rec_limit limits the size of flush groups to avoid
720 * blowing out the UNDO FIFO. This calculation is typically in
721 * the tens of thousands and is designed primarily when small
722 * HAMMER filesystems are created.
724 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
725 if (hammer_debug_general & 0x0001)
726 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
729 * NOTE: Recover stage1 not only handles meta-data recovery, it
730 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
732 error = hammer_recover_stage1(hmp, rootvol);
734 kprintf("Failed to recover HAMMER filesystem on mount\n");
739 * Finish setup now that we have a good root volume.
741 * The top 16 bits of fsid.val[1] is a pfs id.
743 ksnprintf(mp->mnt_stat.f_mntfromname,
744 sizeof(mp->mnt_stat.f_mntfromname), "%s",
745 rootvol->ondisk->vol_name);
746 mp->mnt_stat.f_fsid.val[0] =
747 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
748 mp->mnt_stat.f_fsid.val[1] =
749 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
750 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
752 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
753 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
754 sizeof(mp->mnt_vstat.f_fsid_uuid));
757 * Certain often-modified fields in the root volume are cached in
758 * the hammer_mount structure so we do not have to generate lots
759 * of little UNDO structures for them.
761 * Recopy after recovery. This also has the side effect of
762 * setting our cached undo FIFO's first_offset, which serves to
763 * placemark the FIFO start for the NEXT flush cycle while the
764 * on-disk first_offset represents the LAST flush cycle.
766 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
767 hmp->flush_tid1 = hmp->next_tid;
768 hmp->flush_tid2 = hmp->next_tid;
769 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
770 sizeof(hmp->blockmap));
771 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
773 hammer_flusher_create(hmp);
776 * Locate the root directory using the root cluster's B-Tree as a
777 * starting point. The root directory uses an obj_id of 1.
779 * FUTURE: Leave the root directory cached referenced but unlocked
780 * in hmp->rootvp (need to flush it on unmount).
782 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
786 /*vn_unlock(hmp->rootvp);*/
788 error = hammer_recover_stage2(hmp, rootvol);
791 * If the stage2 recovery fails be sure to clean out all cached
792 * vnodes before throwing away the mount structure or bad things
799 hammer_rel_volume(rootvol, 0);
802 * Cleanup and return.
805 /* called with fs_token held */
808 lwkt_reltoken(&hmp->fs_token);
814 hammer_vfs_unmount(struct mount *mp, int mntflags)
816 hammer_mount_t hmp = (void *)mp->mnt_data;
821 * Clean out the vnodes
823 lwkt_gettoken(&hmp->fs_token);
825 if (mntflags & MNT_FORCE)
827 error = vflush(mp, 0, flags);
830 * Clean up the internal mount structure and related entities. This
834 /* called with fs_token held */
837 lwkt_reltoken(&hmp->fs_token);
843 * Clean up the internal mount structure and disassociate it from the mount.
844 * This may issue I/O.
846 * Called with fs_token held.
849 hammer_free_hmp(struct mount *mp)
851 hammer_mount_t hmp = (void *)mp->mnt_data;
852 hammer_flush_group_t flg;
857 * Flush anything dirty. This won't even run if the
858 * filesystem errored-out.
861 while (hammer_flusher_haswork(hmp)) {
862 hammer_flusher_sync(hmp);
866 kprintf("HAMMER: umount flushing.");
869 tsleep(&dummy, 0, "hmrufl", hz);
872 kprintf("giving up\n");
876 if (count >= 5 && count < 30)
880 * If the mount had a critical error we have to destroy any
881 * remaining inodes before we can finish cleaning up the flusher.
883 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
884 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
885 hammer_destroy_inode_callback, NULL);
889 * There shouldn't be any inodes left now and any left over
890 * flush groups should now be empty.
892 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
893 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
894 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
895 KKASSERT(RB_EMPTY(&flg->flush_tree));
897 kprintf("HAMMER: Warning, flush_group %p was "
898 "not empty on umount!\n", flg);
900 kfree(flg, hmp->m_misc);
904 * We can finally destroy the flusher
906 hammer_flusher_destroy(hmp);
909 * We may have held recovered buffers due to a read-only mount.
910 * These must be discarded.
913 hammer_recover_flush_buffers(hmp, NULL, -1);
916 * Unload buffers and then volumes
918 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
919 hammer_unload_buffer, NULL);
920 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
921 hammer_unload_volume, NULL);
924 mp->mnt_flag &= ~MNT_LOCAL;
926 hammer_destroy_objid_cache(hmp);
927 hammer_destroy_dedup_cache(hmp);
928 if (hmp->dedup_free_cache != NULL) {
929 kfree(hmp->dedup_free_cache, hmp->m_misc);
930 hmp->dedup_free_cache = NULL;
932 kmalloc_destroy(&hmp->m_misc);
933 kmalloc_destroy(&hmp->m_inodes);
934 lwkt_reltoken(&hmp->fs_token);
935 kfree(hmp, M_HAMMER);
939 * Report critical errors. ip may be NULL.
942 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
943 int error, const char *msg)
945 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
947 krateprintf(&hmp->krate,
948 "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
949 hmp->mp->mnt_stat.f_mntfromname,
950 (intmax_t)(ip ? ip->obj_id : -1),
953 if (hmp->ronly == 0) {
954 hmp->ronly = 2; /* special errored read-only mode */
955 hmp->mp->mnt_flag |= MNT_RDONLY;
956 kprintf("HAMMER(%s): Forcing read-only mode\n",
957 hmp->mp->mnt_stat.f_mntfromname);
960 if (hammer_debug_critical)
961 Debugger("Entering debugger");
966 * Obtain a vnode for the specified inode number. An exclusively locked
970 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
971 ino_t ino, struct vnode **vpp)
973 struct hammer_transaction trans;
974 struct hammer_mount *hmp = (void *)mp->mnt_data;
975 struct hammer_inode *ip;
977 u_int32_t localization;
979 lwkt_gettoken(&hmp->fs_token);
980 hammer_simple_transaction(&trans, hmp);
983 * If a directory vnode is supplied (mainly NFS) then we can acquire
984 * the PFS domain from it. Otherwise we would only be able to vget
985 * inodes in the root PFS.
988 localization = HAMMER_DEF_LOCALIZATION +
989 VTOI(dvp)->obj_localization;
991 localization = HAMMER_DEF_LOCALIZATION;
995 * Lookup the requested HAMMER inode. The structure must be
996 * left unlocked while we manipulate the related vnode to avoid
999 ip = hammer_get_inode(&trans, NULL, ino,
1000 hmp->asof, localization,
1005 error = hammer_get_vnode(ip, vpp);
1006 hammer_rel_inode(ip, 0);
1008 hammer_done_transaction(&trans);
1009 lwkt_reltoken(&hmp->fs_token);
1014 * Return the root vnode for the filesystem.
1016 * HAMMER stores the root vnode in the hammer_mount structure so
1017 * getting it is easy.
1020 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1024 error = hammer_vfs_vget(mp, NULL, 1, vpp);
1029 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1031 struct hammer_mount *hmp = (void *)mp->mnt_data;
1032 hammer_volume_t volume;
1033 hammer_volume_ondisk_t ondisk;
1038 lwkt_gettoken(&hmp->fs_token);
1039 volume = hammer_get_root_volume(hmp, &error);
1041 lwkt_reltoken(&hmp->fs_token);
1044 ondisk = volume->ondisk;
1049 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1050 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1051 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1052 hammer_rel_volume(volume, 0);
1054 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1055 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1056 if (mp->mnt_stat.f_files < 0)
1057 mp->mnt_stat.f_files = 0;
1059 *sbp = mp->mnt_stat;
1060 lwkt_reltoken(&hmp->fs_token);
1065 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1067 struct hammer_mount *hmp = (void *)mp->mnt_data;
1068 hammer_volume_t volume;
1069 hammer_volume_ondisk_t ondisk;
1074 lwkt_gettoken(&hmp->fs_token);
1075 volume = hammer_get_root_volume(hmp, &error);
1077 lwkt_reltoken(&hmp->fs_token);
1080 ondisk = volume->ondisk;
1085 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1086 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1087 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1088 hammer_rel_volume(volume, 0);
1090 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1091 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1092 if (mp->mnt_vstat.f_files < 0)
1093 mp->mnt_vstat.f_files = 0;
1094 *sbp = mp->mnt_vstat;
1095 lwkt_reltoken(&hmp->fs_token);
1100 * Sync the filesystem. Currently we have to run it twice, the second
1101 * one will advance the undo start index to the end index, so if a crash
1102 * occurs no undos will be run on mount.
1104 * We do not sync the filesystem if we are called from a panic. If we did
1105 * we might end up blowing up a sync that was already in progress.
1108 hammer_vfs_sync(struct mount *mp, int waitfor)
1110 struct hammer_mount *hmp = (void *)mp->mnt_data;
1113 lwkt_gettoken(&hmp->fs_token);
1114 if (panicstr == NULL) {
1115 error = hammer_sync_hmp(hmp, waitfor);
1119 lwkt_reltoken(&hmp->fs_token);
1124 * Convert a vnode to a file handle.
1126 * Accesses read-only fields on already-referenced structures so
1127 * no token is needed.
1130 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1134 KKASSERT(MAXFIDSZ >= 16);
1136 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1137 fhp->fid_ext = ip->obj_localization >> 16;
1138 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1139 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1145 * Convert a file handle back to a vnode.
1147 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1151 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1152 struct fid *fhp, struct vnode **vpp)
1154 hammer_mount_t hmp = (void *)mp->mnt_data;
1155 struct hammer_transaction trans;
1156 struct hammer_inode *ip;
1157 struct hammer_inode_info info;
1159 u_int32_t localization;
1161 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1162 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1164 localization = VTOI(rootvp)->obj_localization;
1166 localization = (u_int32_t)fhp->fid_ext << 16;
1168 lwkt_gettoken(&hmp->fs_token);
1169 hammer_simple_transaction(&trans, hmp);
1172 * Get/allocate the hammer_inode structure. The structure must be
1173 * unlocked while we manipulate the related vnode to avoid a
1176 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1177 info.obj_asof, localization, 0, &error);
1179 error = hammer_get_vnode(ip, vpp);
1180 hammer_rel_inode(ip, 0);
1184 hammer_done_transaction(&trans);
1185 lwkt_reltoken(&hmp->fs_token);
1190 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1191 int *exflagsp, struct ucred **credanonp)
1193 hammer_mount_t hmp = (void *)mp->mnt_data;
1197 lwkt_gettoken(&hmp->fs_token);
1198 np = vfs_export_lookup(mp, &hmp->export, nam);
1200 *exflagsp = np->netc_exflags;
1201 *credanonp = &np->netc_anon;
1206 lwkt_reltoken(&hmp->fs_token);
1212 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1214 hammer_mount_t hmp = (void *)mp->mnt_data;
1217 lwkt_gettoken(&hmp->fs_token);
1220 case MOUNTCTL_SET_EXPORT:
1221 error = vfs_export(mp, &hmp->export, export);
1227 lwkt_reltoken(&hmp->fs_token);